[BUG] `dynamic_partition_pruning::read_table` errors on single-file Parquet datasets

[charlesbluca]

What happened:
While playing around with hooking dask-sql into Coiled's benchmarks, I noticed some issues around DPP with test_query_3:

pyo3_runtime.PanicException: called `Result::unwrap()` on an `Err` value: Os { code: 20, kind: NotADirectory, message: "Not a directory" }

Think this is because dynamic_partition_pruning::read_table assumes we're working with a directory of chunked parquet files and doesn't have handling for the case where we have a single parquet file:

dask-sql/src/sql/optimizer/dynamic_partition_pruning.rs

Line 488 in 2f068f9

let paths = fs::read_dir(tables.get(&table_string).unwrap().filepath.clone()).unwrap();

What you expected to happen:
In general, I would expect this to emit a warning and skip DPP rather than bubble up to an error, though I don't think it should be too difficult to modify the handling of tables for the single file case? cc @sarahyurick

Environment:

• dask-sql version: 2023.10.1
• Python version: 3.9
• Operating System: ubuntu20.04
• Install method (conda, pip, source): conda